NEUROLOGICAL FOUNDATIONS OF COGNITIVE NEUROSCIENCE - PART 7 pot

Michael P Alexander in the left lateral frontal lobe TCMA (or Luria’s dynamic aphasia) represents the range of aphasic disorders in which the fundamental processes— semantics, phonology, articulation, grammar, and concatenation—are normal, but the utilization of them is impaired Clinical, imaging, and cognitive neuroscience investigations in the past 25 years have sharpened our understanding of TCMA and clarified its neural and psychological components, although Luria’s basic characterizations remain fundamental even to modern concepts Lesion specificity has been clarified The roles of different regions of the frontal lobes in discrete aspects of language are better understood Insights from other domains of cognitive neuroscience have illuminated the mechanisms of planning and intention in speech Lesion–Anatomy Correlations in TCMA Any analysis of the language disorders due to frontal lesions must begin with Broca’s aphasia The eponymous area is usually marked with a “B” and lies over the frontal operculum, roughly Brodmann areas 44 and 45; sometimes it includes the lower motor cortex (area 4) and the anterior, superior insular cortex continuous with the inferior opercular surface Damage restricted to these areas produces a somewhat variable clinical picture, sometimes called “Broca’s area aphasia” (Mohr et al., 1978) In the acute phase, these patients have more similarities than differences They are often briefly mute, then show effortful speech with articulation and prosody impairments, reduced phrase length, syntax errors, and mixed paraphasias, all variably but modestly benefited by repetition Thus, Broca’s area lesions produce acute Broca’s aphasia In the chronic phase, these patients diverge along several paths (Alexander, Naeser, & Palumbo, 1990) Lesions centered in the posterior operculum and the lower motor cortex are likely to cause persistent articulation and prosody impairments, with rapid recovery of lengthy, grammatical utterances Lesions centered in the anterior superior operculum 168 are likely to produce persistent truncation of utterances, although without much overt grammatical impairment, with rapid recovery of articulation and prosody and rapid normalization of repetition and recitation Thus, viewed from the postacute perspective, Broca’s area lesions damage two adjacent, perhaps overlapping, neural systems, one fundamentally for motor control of speech and one for realization of lengthy, complex utterances Broca’s area lesions not produce lasting Broca’s aphasia Freedman and colleagues (Freedman, Alexander, & Naeser, 1984) analyzed a large number of patients in the postacute stage that met a standard clinical definition of TCMA (Goodglass & Kaplan, 1983) More than one lesion site was identified Some patients had damage to the frontal operculum, including the anterior portions of Broca’s area Some had damage to more dorsolateral midfrontal regions, which often projected into white matter Some had damage only to the deep white matter including or adjacent to and above the head of the caudate nucleus Some had large capsulostriatal lesions reaching up to the head of the caudate nucleus and the adjacent white matter Some had medial frontal damage, including the supplementary motor area (SMA) Earlier descriptions of aphasia after infarctions of the left anterior cerebral artery (ACA) territory or associated with parasagittal tumors had already established that large medial frontal lesions produced a speech and language impairment (Critchley, 1930) Mutism, paucity of speech, and repetitive utterances were described Several reports in the 1970s (Von Stockert, 1974; Rubens, 1976) (Masdeu, Schoene, & Funkenstein, 1978) and 1980s (Alexander & Schmitt, 1980) defined the evolution of aphasia with left medial frontal lesions: initial mutism for hours to weeks and then gradual recovery of lengthy, fluent output, with preserved repetition and recitation In the report by Freedman and colleagues, a detailed assessment of the variation in postacute language impairment associated with left lateral frontal damage revealed the important anteriorposterior divergence of roles within the frontal Transcortical Motor Aphasia cortex (Freedman et al., 1984) The posterior portions are essential for articulation; the anterior portions are essential for some aspect of generative language—complex sentences, narratives, etc.—but are unimportant for externally driven language—repetition, naming, oral reading, and short responses There is considerable controversy about the so-called “subcortical aphasias,” particularly those associated with left capsulostriatal lesions An analysis of absolute cortical perfusion and of the extent and location of carotid obstructive disease suggests to some investigators that aphasia is due to cortical hypoperfusion, causing microscopic cortical neuronal injury (Olsen, Bruhn, & Oberg, 1986) (Nadeau & Crosson, 1995) In this view, the subcortical lesion is irrelevant With numerous collaborators, I have proposed a different mechanism for aphasia (Alexander, Naeser et al., 1987) Most structures within capsulostriatal lesions are, in fact, irrelevant to aphasia Lesions in the putamen, the globus pallidus ventral anterior limb internal capsule (ALIC), or most of the paraventricular white matter (PVWM) not appear to affect language Lesions in the dorsal ALIC, the dorsal head of the caudate nucleus and the anterior PVWM, on the other hand, are associated with a mild generative aphasia, i.e., TCMA, in the postacute period (Mega & Alexander, 1994) These patients also often have severe articulatory impairment (descending corticobulbar pathways), hypophonia (putamen), and hemiparesis (corticospinal pathways) None of these are pertinent to aphasia; the aphasia diagnosis is independent of the neurological findings (Alexander et al., 1987) Spontaneous (hypertensive) hemorrhages in capsulostriatal territories produce a more severe initial aphasia and a broader range of aphasias in the postacute period because a dissection of a hemorrhage can produce idiosyncratic lesion extensions (D’Esposito & Alexander, 1995) The “core syndrome” of mild TCMA after lesions in caudate or anterior white matter is maintained Consolidation of these disparate observations is possible Damage to the medial frontal cortex, 169 including the SMA and anterior cingulate gyrus (ACG), produces akinetic mutism (Freemon, 1971) The akinesia, including akinesia of the speech apparatus (i.e., mutism), is due to the loss of ascending cortical dopaminergic input (Lindvall, Bjorkland, Moorc, Steneui, 1974) Thus, the progressive aphasia commonly associated with progressive supranuclear palsy (PSP) is dynamic aphasia or TCMA, although it is often embedded in more pervasive activation and executive impairments (Esmonde, Giles et al., 1996) The SMA (Jürgens, 1984) and ACG (Baleydier & Mauguiere, 1980) have interesting connectivity principles Afferents are received from all sensory association cortices and potently from dopaminergic brainstem nuclei, but efferents are bilateral to all frontal regions and to the striatum Thus, processed sensory information converges with subcortical drive and activation mechanisms The resultant output from the SMA and ACG is the activation transformer of the brain Medial structures provide the drive for continued sustained movement and cognition Projections through anterior PVWM regions and to the caudate nucleus carry this activation to the lateral frontal regions, converging on the left frontal operculum for speech (Alexander et al., 1987) Lesions anywhere in this system will damage drive, activation, and generative capacities, producing truncated, unelaborated language Thus, damage to this efferent, bilateral medial to left lateral frontal system is the foundation for the impairment observed in “intention” to speak Simple responses, recitation, repetition, even naming require much less generative effort; thus they are preserved The posterior operculum, in turn, organizes motor programs of speech Modern Notions of Dynamic Aphasia Recent investigators have analyzed the cognitive and linguistic impairments that might underlie the planning and supervisory deficits in TCMA by focusing on dynamic aphasia, the cleanest exemplar of TCMA Some extrapolation from functional Michael P Alexander neuroimaging studies in normal subjects also illuminates this issue These investigations have attempted to specify more precisely the testable deficits that make up the generality of “planning.” The most carefully analyzed single case reports of dynamic aphasia meet clinical criteria for TCMA with left frontal lesions Costello and Warrington (1989) demonstrated that their patient was unable to produce a conceptual structure for an utterance prior to any implementation of syntactic options for expression and prior to actual sentence production Robinson et al observed that their patient was unable to select propositional language when the communication context provided little constraint or prompting (Robinson, Blair, & Cipolotti, 1998) When there were numerous possible utterances and constructions, the patient was impaired When context defined a response, language was normal Thompson-Schill et al have shown the same type of deficit at the single-word level in patients with lesions that included the left posterior frontal regions (Thompson-Schill, Swick, Farah, D’Esposito, Kan, & Knight, 1998) Language activation studies with positron emission tomography (PET) (Petersen, Fox, Posner, Mintun, & Raichle, 1989) or functional magnetic resonance imaging (fMRI) (Desmond, 1995) have long demonstrated that the left frontal opercular area is activated in tasks of semantic generation, such as naming a verb that is associated with a given noun This activation is not just associated with semantic retrieval, but depends as much on selection of an item from a range of retrieved choices (Thompson-Schill, D’Esposito, Aguirre, & Farah, 1997) Patients with posterior frontal lesions have difficulty with verb generation in proportion to the number of choices available to them (Thompson-Schill et al., 1998) Nadeau (1988) analyzed the syntactic constructions of two patients with large left lateral frontal lesions He demonstrated that word choice and grammar within a sentence can be intact when the syntactic frame selected for the overall response is defective In a PET study of memory retrieval in normal subjects, Fletcher et al observed a distinction in left frontal activation, depending on the relationship of 170 word pairs to be retrieved Thus, retrieval and production of verbal material that was highly probably linked, whether imageable (arm-muscle) or not (happiness-love), produced little left frontal activation When retrieval required construction of novel links between unrelated word pairs, even if they were highly imageable individually (hurricanepuppy), there was marked left lateral frontal activation (Fletcher, Shallice, Frith, Frackowiak, & Dolan, 1996) The authors remarked on the similarity of this finding to the difficulty that patients with left frontal lesions and dynamic aphasia have producing responses that are not highly connected semantically All of these potential explanations for dynamic aphasia revolve around impaired language planning when the context of the utterance does not immediately guide output Whether at the word or sentence level (or even at the discourse level; see the following discussion), this planning and selection problem appears fundamental to frontal aphasias When numerous responses are possible, when word and syntax selections are not constrained, when social context does not restrict the form that utterances might take, the left frontal region is critical for selection and execution of a particular response strategy This is action planning in the domain of language Discourse Discourse is the production of structured complex output (Chapman, Culhane et al 1992) During development, humans learn rules and accepted procedures for discourse and in parallel, they learn how and when to use these procedures (Chapman, Culhane et al., 1992) They learn a “theory of mind,” that is, the capacity to place themselves in a listener’s mind to estimate what knowledge or expectations or emotions the listener might bring to an interaction (Stone, Baron-Cohen, & Knight, 1998; Gallagher, Happe, Brunswick, Fletcher, Frith, & Frith, 2000) They learn the context and constraints for the use of discourse They learn their culture’s rules, styles, and strategies for discourse Transcortical Motor Aphasia Some forms of discourse are highly rule bound: pleading a court case, structuring a medical report, writing a book chapter, and telling some types of jokes Discourse can be narrative (telling a story) or procedural (relating a recipe, teaching car repair) or a mixture of both (teaching biology) The forms of discourse have rules of construction (story grammar), rules of coherence (using intelligible references), rules of indirection, etc Prefrontal lesions produce impairments in discourse (Kaczmarek, 1984; Chapman et al., 1992) The discourse errors of left prefrontal lesions are mostly simplifications (Novoa & Ardila, 1987) There is a reduction in variation of sentence structure and a tendency to repeat sentence forms There is a reduction in the number of relevant themes and concepts recruited to fill out a narrative; thus reference within a narrative is often incomplete The boundary between dynamic aphasia and defective discourse is not fixed Patients with dynamic aphasia use simple and unelaborated sentence forms and tend to repeat a few sentence structures There are clearly nested levels of impairment in the recruitment of the elements of complex language Thus far this review has only dealt with left frontal lesions At the level of discourse, right prefrontal injury may also disrupt communication (Novoa & Ardila, 1987) The limited evidence suggests that right prefrontal lesions reduce organization and monitoring, allowing the tangential, unrelated, and at times inappropriate and in some cases, frankly confabulatory narratives characteristic of right frontal damage Production of complex language presupposes intact fundamental language processes—phonetics, phonology, semantics, and grammar Using those preserved functions, a large group of interrelated operations must unfold to produce complex language The operations include selection of discourse intention and form, allowing for shared knowledge with the listener; selection of syntactic procedures that fit the intended communication; and selection from the many options of the precise lexical elements that express the intentions and fit the syntax How all of this unfolds online is beyond the abili- 171 ties of this writer and is a complex, vital issue in cognitive science (Levelt, 1989), but at the “offline” level of impairments due to frontal injury, we return to action planning Action Planning Action planning has been evaluated in patients with neurological damage The models for action planning vary somewhat (Shallice, 1982; Schwartz, Reed, Montgomery, Palmer, & Mayer, 1991) All appear to suppose that experience has taught everyone a wide variety of simple actions (pouring, cutting, untwisting, etc.) and of possible assemblies of those actions to achieve certain goals (fixing coffee, making a sandwich, etc.) When some actions are frequently combined in an unvarying manner, then the resulting practiced complex action may become a unit of action of its own (eating breakfast, getting dressed) Across life’s experiences, a large repertoire of simple and combined actions become proceduralized, that is, produced as a whole without explicit conscious direction As the complexity of action increases and as the possible order of recruitment of subparts of the action (schemas) becomes less fixed, more explicit conscious direction is required to select and assemble the parts into an intended whole, delaying or holding some actions, inhibiting others, and monitoring progress to the goal (intention) Deficits in action planning have been studied with simple everyday behaviors, such as eating breakfast (Schwartz et al., 1991), and with more complex behaviors, such as shopping (Shallice & Burgess, 1991) TCMA, at least dynamic aphasia, and discourse deficits are action planning failures in language Patients cannot generate a plan or subplans, select from among alternative plans, or maintain an initial selection without contamination from other activated possible plans; nor can they keep track of how the several selected plans are progressing This assembly and planning function operates at numerous levels that appear to have anterior-posterior arrangements in the left frontal lobe (Sirigu, Cohen Michael P Alexander et al 1998) In the posterior ventrolateral frontal lobe, deficits may be at the level of word activation and selection (Thompson-Schill et al., 1998) Thus, language is quite restricted whenever the response is not prompted by words in the question or some other externality With lesions of the dorsolateral frontal lobe, deficits may be at the level of syntactic selection (Costello & Warrington, 1989) Language is restricted whenever a novel sentence structure must be generated and, in default, any provided sentence may be pirated, at least in part, to carry the response; thus echolalia and perseveration With prefrontal lesions, deficits may be at the discourse level Language is produced and word selection proceeds, but the organization of plans for complex action (discourse) is impaired There may be reliance on a few syntactic forms to carry the communication load and great difficulty generating new syntactic or narrative structures 172 the frontal language disorders The essential frontal language disorder is TCMA The deficits in TCMA are a mixture of delayed initiation (even mutism), impaired lexical selection, and reduced capacity to generate unconstrained syntactic forms The prototypical lesions are in the left lateral frontal cortex, including much of the classic Broca’s area, or in subcortical structures, including white matter projections and dorsal caudate nucleus The two fundamental factors that underlie defective language production after a left frontal lobe injury are intention and planning Intention deficits are due to damage to medial frontal structures, their afferent projections, or their efferent convergence in left lateral frontal regions, probably quite diffusely Planning deficits are due to damage to the left lateral frontal lobe, again rather diffusely, with interleaved impairments in planning extending from the level of word selection to syntax selection to discourse construction roughly correlating with a posterior-topolar progression of frontal lesions Conclusions and Future Directions Dynamic aphasia appears to be an ideal substrate for analyzing the elements of action planning Mapping the conceptual framework of action plans on to language production should be a path to a clearer understanding of both If the elements of TCMA or dynamic aphasia are well defined now, methodologies for treatments are not Is it possible to re-train the use of complex syntax or discourse? Can patients learn substitutions and compensatory rules or must complex language be rehearsed and practiced in a natural context? Can planning be taught offline with picture and story arrangement tasks or can it only be relearned in the process of speaking? Does dopaminergic deficiency actually underlie any component of the language deficit (Sabe, Salvarezza, Garcia Cuerva, Leiguarda, & Starkstein, 1995) or is it only relevant to the more pervasive akinetic mutism syndromes (Ross & Stewart, 1981)? The progress from Goldstein to Shallice is palpable, but as yet of little benefit to patients There are embedded impairments in action planning for language that in their interactions make up References Alexander, M P (1997) Aphasia: Clinical and anatomic aspects In T E Feinberg, & M J Farah (Eds.), Behavioral neurology and neuropsychology (pp 133–149) New York: McGraw-Hill Alexander, M P., Naeser, M A., & Palumbo, C L (1987) Correlations of subcortical CT lesion sites and aphasia profiles Brain, 110, 961–991 Alexander, M P., Naeser, M A et al (1990) Broca’s area aphasia Neurology, 40, 353–362 Alexander, M P., & Schmitt, M A (1980) The aphasia syndrome of stroke in the left anterior cerebral artery territory Archives of Neurology, 37, 97–100 Baleydier, C., & Mauguiere, F (1980) The duality of the cingulate gyrus in monkey: Neuroanatomical study and functional hypothesis Brain, 103, 525–554 Chapman, S B., Culhane, K A., Levin, H S., Harward, H., Mendelsohn, D., Ewing-Cobbs, L., Fletcher, J M., & Bruce, D (1992) Narrative discourse after closed head injury in children and adolescents Brain and Language, 43, 42–65 Transcortical Motor Aphasia Costello, A L., & Warrington, E K (1989) Dynamic aphasia The selective impairment of verbal planning Cortex, 25, 103–114 Critchley, M (1930) Anterior cerebral artery and its syndromes Brain, 53, 120–165 D’Esposito, M., & Alexander, M P (1995) Subcortical aphasia: Distinct profiles following left putaminal hemorrhages Neurology, 45, 33–37 Desmond, J E., Sum, J M., Wagner, A D., Demb, J B., Shear, P K., Glover, G H., Gabrieli, J D., & Morrell, M J (1995) Functional MRI measurement of language lateralization in Wada-tested patients Brain, 118, 1411–1419 Esmonde, T., Giles, E., Xuereb, J., & Hodges, J (1996) Progressive supranuclear palsy presenting with dynamic aphasia Journal of Neurology, Neurosurgery and Psychiatry, 60, 403–410 Fletcher, P C., Shallice, T., Frith, C D., Frackowiak, R S., & Dolan, R J (1996) Brain activity during memory retrieval: The influence of imagery and semantic cueing Brain, 119, 1587–1596 Freedman, M., Alexander, M P., & Naeser, M A (1984) Anatomic basis of transcortical motor aphasia Neurology, 34, 409–417 Freemon, F R (1971) Akinetic mutism and bilateral anterior cerebral artery occlusion Journal of Neurology, Neurosurgery and Psychiatry, 34, 693–698 Gallagher, H L., Happe, F., Brunswick, N., Fletcher, P C., Frith, U., & Frith, C D (2000) Reading the mind in cartoons and stories: An fMRI study of “theory of mind” in verbal and nonverbal tasks Neuropsychologia, 38, 11–21 Goldstein, K (1948) Language and language disorders New York: Grune & Stratton Goodglass, H (1993) Understanding aphasia San Diego: Academic Press Goodglass, H., & Kaplan, E (1983) The assessment of aphasia and related disorders Philadelphia: Lea & Febiger Jürgens, U (1984) The efferent and afferent connections of the supplementary motor area Brain Research, 300, 63–81 Kaczmarek, B L J (1984) Neurolinguistic analysis of verbal utterances in patients with focal lesions of frontal lobes Brain and Language, 21, 52–58 Levelt, W J M (1989) Speaking: From intention to articulation Cambridge, MA: MIT Press 173 Lichtheim, L (1885) On aphasia Brain, 7, 433–484 Lindvall, O., Bjorkland, A., Moore, R Y., & Stenevi, U (1974) Mesencephalic dopamine neurons projecting to neocortex Brain Research, 81, 325–331 Luria, A R (1973) The working brain New York: Basic Books Luria, A R., & Tsevtkova, L S (1967) Towards the mechanism of “dynamic aphasia” Acta Neurologica Psychiatrica Belgica, 67, 1045–1067 Masdeu, J C., Schoene, W C., & Funkenstein, H (1978) Aphasia following infarction of the left supplementary motor area Neurology, 28, 1220–1223 Mega, M S., & Alexander, M P (1994) The core profile of subcortical aphasia Neurology, 44, 1824–1829 Mohr, J P., Pessin, M et al (1978) Broca aphasia: Pathologic and clinical aspects Neurology, 28, 311–324 Nadeau, S (1988) Impaired grammar with normal fluency and phonology Brain, 111, 1111–1137 Nadeau, S., & Crosson, B (1995) Subcortical aphasia Brain and Language, 58, 355–402 Novoa, O P., & Ardila, A (1987) Linguistic abilities in patients with prefrontal damage Brain and Language, 30, 206–225 Petersen, S E., Fox, P T., Posner, M I., Mintun, M., & Raichle, M E (1989) Positron emission tomographic studies of the processing of single words Journal of Cognitive Neuroscience, 1, 153–170 Robinson, G., Blair, J., & Cipolotti, L (1998) Dynamic aphasia: An inability to select between competing verbal responses? Brain, 121, 77–89 Ross, E D., & Stewart, R M (1981) Akinetic mutism from hypothalamic damage: Successful treatment with dopamine agonists Neurology, 31, 1435–1439 Rubens, A B (1976) Transcortical motor aphasia Studies in Neurolinguistics, 1, 293–306 Sabe, L., Salvarezza, F., Garcia Cuerva, A., Leiguarda, R., & Starkstein, S (1995) A randomized, double-blind, placebo-controlled study of bromocriptine in nonfluent aphasia Neurology, 45, 2272–2274 Schwartz, M F., Reed, E S., Montgomerv, M., Palmer, C., & Mayer, N H (1991) The quantitative description of action disorganization after brain damage: A case study Cognitive Neuropsychology, 8, 381–414 Shallice, T (1982) Specific impairments of planning Philosophical Transactions of the Royal Society of London, 298, 199–209 Michael P Alexander Shallice, T., & Burgess, P W (1991) Deficits in strategy application following frontal lobe damage in man Brain, 114, 727–741 Sirigu, A., Cohen, L., Zalla, T., Pradat-Diehl, P., Van Eechout, P., Grafman, J., & Agid, Y (1998) Distinct frontal regions for processing sentence syntax and story grammar Cortex, 34, 771–778 Olsen, T S., Bruhn, P., & Oberg, R G (1986) Cortical hypoperfusion as a possible cause of “subcortical aphasia.” Brain, 109, 393–410 Stone, V E., Baron-Cohen, S., & Knight, R T (1998) Frontal lobe contributions to theory of mind Journal of Cognitive Neuroscience, 10, 640–656 Thompson-Schill, S L., D’Esposito, M., Aguirre, G K., & Farah, M J (1997) Role of left inferior prefrontal cortex in retrieval of semantic knowledge: A reevaluation Proceedings of the National Academy of Sciences U.S.A., 94, 14792–14797 Thompson-Schill, S L., Swick, D., Farah, M J., D’Esposito, M., Kan, I P., & Knight, R T (1998) Verb generation in patients with focal frontal lesions: A neuropsychological test of neuroimaging findings Proceedings of the National Academy of Sciences U.S.A., 95, 15855–15860 Von Stockert, T R (1974) Aphasia sine aphasie Brain and Language, 1, 277–282 174 Wernicke Aphasia: A Disorder of Central Language Processing Jeffrey R Binder Case Report Patient H.K is a 75-year-old, right-handed woman with mild hypertension who suddenly developed language difficulty and right hemiparesis Prior to this, she had been healthy, living alone and managing her own affairs Hemiparesis was confined to the right face and hand and resolved within 24 hours Persistent language deficits observed during the acute hospitalization included poor naming of objects, difficulty producing understandable words in speech, and impaired understanding of commands and questions A computed tomography (CT) scan obtained on the third day after onset showed an acute infarction in the territory of the left middle cerebral artery, affecting posterior temporal and parietal regions She was discharged home after week Although she was able to perform all necessary activities such as shopping, cooking, and cleaning, persistent communication deficits made social interactions difficult and embarassing Initial Examination When examined in more detail weeks after onset, the patient was alert and able to write her name, the date, and the name of the hospital She was calm and attentive, always attempting to understand and comply with what was requested of her She spoke frequently and with fluent, well-articulated production of phonemes Her sentences were of normal length and prosody Spontaneously uttered words were mostly recognizable except for occasional neologisms (nonwords) Her word output consisted almost entirely of familiar combinations of closedclass words (articles, prepositions, pronouns) and common verbs, with relatively little noun content The following is a transcription of her description of the Cookie Theft Picture from the Boston Diagnostic Aphasia Evaluation (BDAE) (Goodglass & Kaplan, 1972): “What has he got here? That that’s coming right over there, I’ll tell you that This is the the conner? the bonner falling down here And that’s the boy going to getting with it over there She’s got this washering it’s upside, and down She’d doing the the fixing it, the plape? the plate, that she’s got it there And on it, the girl’s sort of upside Is that about? Anything else I’m missing, if it’s down, that I wouldn’t know?” Verbal and phonemic paraphasias were more common in tasks requiring production of specific words, such as naming, repeating, and reading She was unable to name correctly any presented objects, pictures, or colors, but produced neologistic utterances for many of these (“hudder” for hammer, “remp” for red), as well as occasional semantically related words (“dog” for horse) Her responses were characterized by repeated attempts and successively closer phonemic approximations to the target word (“fleeth, fleth, fleether, fleather” for feather) Naming of numbers and letters was sometimes correct, and more often than with objects resulted in semantic substitution of other items in the same category Strikingly, she was often able to write correctly the names of objects she was unable to pronounce After failing to name orally six object pictures from the BDAE (glove, key, cactus, chair, feather, hammock), she succeeded in writing four of these correctly (cactus, chair, feather, hammock) and wrote a semantically related word for the others (“hand” for glove and “lock” for key) Repetition was severely defective for all stimuli Even after correctly writing the names for objects, she was unable to repeat these names aloud after hearing the examiner and simultaneously looking at the name she had just written Errors in repetition and reading aloud were almost entirely phonemic paraphasias She was often able to write to dictation familiar nouns she could not repeat aloud (dog, cat, horse, hand, ear, nose), but was unable to this with less common words (sheep, goat, trout, jaw, chin, knee) She was unable to write a simple sentence to dictation (For “A boy had a dog,” she wrote with some hesitation “He and aswer”) She followed simple oral commands given without accompanying gestures (“close eyes,” Jeffrey R Binder “open mouth,” “smile,” “stand up”) in approximately half of the trials, possibly inferring some of the meaning from context She was unable to follow less likely commands (“look left,” “lick lips,” “clench jaw,” “lean back”) or multicomponent commands Simple questions containing five to seven words (“Did you eat lunch today?” “How did you get here?”) evoked fluent, empty responses with no apparent relationship to the question On auditoryvisual matching tasks using six to eight-item visual arrays, she was able to point to named objects, words, and letters with 100% accuracy, indicating preservation of some auditory comprehension for single words She understood written commands and questions no better than the auditory versions The remainder of the neurological examination was normal, including tests for visual neglect, visual field, and other cranial nerve tests, motor and sensory examination, and cerebellar and gait testing Structural Magnetic Resonance Imaging High-resolution, T1-weighted magnetic resonance images (MRI) (voxel size = mm3) were obtained 14 months postonset (figure 9.1) A large region of 176 encephalomalacia was observed in the posterior left hemisphere Damaged areas included most of Heschl’s gyrus (HG) and the planum temporale (PT), the superior temporal gyrus (STG) and superior temporal sulcus (STS) lateral and ventral to HG, and the dorsal aspect of the posterior middle temporal gyrus (MTG) Left parietal lobe damage affected the entire supramarginal gyrus (SMG) except for a thin ribbon of preserved cortex along the intraparietal sulcus, and approximately the anterior two-thirds of the angular gyrus (AG) Subcortical white matter was destroyed in these gyri, while deep periventricular white matter was spared Subsequent Course Severe aphasic deficits have persisted over years of follow-up, although the patient remains able to manage all daily necessities of living Spontaneous speech remains fluent, with relatively little noun or adjective content Oral confrontation naming has improved modestly, so that the patient succeeds in a small proportion of trials, but with frequent phonemic paraphasias and successive approximations to the target (“coxis, caxis, coctis, cactus” for cactus) Written naming is consistently superior to Figure 9.1 A T1-weighted MRI in patient H.K In the top row are serial coronal slices through the posterior perisylvian region, taken at 10-mm intervals and arranged anterior to posterior The left hemisphere is on the reader’s right The bottom row shows serial sagittal slices through the left hemisphere at 7-mm intervals The position of the coronal slices is indicated by the vertical lines in the third image Wernicke Aphasia 177 oral naming, and writing to dictation remains notably better than oral repetition The patient has spontaneously developed a strategy of writing down or spelling aloud what she is trying to say when listeners not appear to understand At months postonset, she produced the following transcription of several simple sentences she was unable to repeat orally: I cannot go home without my dog (The boy werit that the I home.) Then the boy began to cry (He carire cried.) The ability to carry out simple oral commands is now more consistent, whereas comprehension of multistep commands and simple questions not related to the immediate context remains severely deficient in both auditory and visual modalities Auditory Stimulus (Patient’s Transcription) Clinical Description of Wernicke Aphasia A boy had a dog (A boy and girl found dog.) The dog ran into the woods (The dogs run into the woods.) The boy ran after the dog (The boy ran away the dog.) He wanted the dog to go home (The boys run and the dog is all home.) But the dog would not go home (The bog isn’t home.) The little boy said (The little boy was) Like the other aphasias, Wernicke aphasia is a syndrome complex composed of several distinct signs (table 9.1) The central characteristic is a disturbance of language comprehension, manifested by incorrect or unexpected responses to spoken commands and other language stimuli In the acute stage, this deficit may be so severe as to seem to involve more than language alone, the patient often appearing to show no reaction to verbal input from Table 9.1 Characteristic clinical features of Wernicke aphasia and several related syndromes Clinical Syndromes Tasks Wernicke aphasia Transcortical sensory aphasia Pure word deafness Conduction aphasia Comprehension Auditory verbal Written Impaired Impaired Impaired Impaired Impaired Normal Normal Normal Phonemic + verbal Verbal > phonemic Phonemic Phonemic Paraphasic and/or anomic Paraphasic and/or anomic Paraphasic Paraphasic Paraphasic and/or anomic Paraphasic and/or anomic Normal Paraphasic or alexic ± Paraphasic ± Paraphasic Paraphasic ± Paraphasic Paraphasic Paraphasic Paraphasic Paraphasic Paragraphic/anomic Paragraphic/anomic Paragraphic Paragraphic/anomic Paragraphic/anomic ± Lexical agraphia Normal Normal Paragraphic Normal Normal ± Phonological agraphia Production Error type Speech Propositional Naming Repetition Reading aloud Writing Propositional Naming Dictation Wernicke Aphasia refer to concepts Mappings and develop as we learn to read, enabling reading comprehension and reading aloud Pathway 6, though not strictly necessary for reading comprehension, probably develops because of the quasi-regular correspondence between graphemes and phonemes and may be encouraged by teaching methods that emphasize “sounding out” and reading aloud Finally, mappings and permit concepts (as in propositional writing) or heard phonemes (as in writing to dictation) to be translated into written form One important class of intermediary code postulated to play a role in these mapping processes is the whole-word or lexical representation For example, the mapping from input phonemes to semantics is often envisioned as involving an intermediate “phonological input lexicon” composed of whole-word representations that become active as a result of input from appropriate representations in the input phoneme level and send activation, in turn, to the semantic level Such whole-word representations correspond closely to Wernicke’s concept of word-sound images In Wernicke’s model, the same center for word-sound images participates in the mappings marked 1, 3, 4, 5, 6, and in figure 9.2 As we will see, modern neurolinguistic studies provide evidence for at least a partial separation of these pathways As a result of this evidence, there has flourished the idea of a separate phonological input lexicon mediating mapping 3, a phonological output lexicon mediating mapping 4, an orthographic input lexicon mediating mapping 5, and an orthographic output lexicon mediating mapping Precisely how these mappings are actually accomplished is another question, one not addressed at all by the classic Wernicke–Lichtheim model of language processing nor by many recent modular models composed entirely of boxes and arrows How, for example, can there be transformations between entities so dissimilar as phonemes and concepts? At the root of this problem is the fact that there exists no regular relationship between a word’s sounds and its meaning (e.g., words as dif- 183 ferent in meaning as cat, cot, coat, and cut nevertheless sound very similar); the mapping between phonemes and semantics is essentially arbitrary The idea that a lexicon of word representations links phonemes to meanings reflects our intuition that something is needed to mediate between these very different kinds of information Explicit neural network simulations of these same mappings, explored in some detail over the past 20 years, support this intuition by demonstrating that arbitrary mappings of this sort can only be accomplished by adding an intermediary (or hidden) representational level between the input and output levels.2 As we will see, the notion of intermediate representational levels is central to understanding both the pathophysiology of aphasia and the nature of the activations observed in functional imaging experiments Figure 9.3 shows a somewhat more realistic language-processing architecture complete with intermediate representational levels supporting arbitrary and quasi-regular mappings The figure makes clear the parallel between these intermediate representations and the “lexicons” of cognitive neuropsychology The implication of this comparison is that models postulating lexicons with whole-word representations are but one possible version of a more general architecture based on intermediate representations In contrast to the whole-word model, neural network simulations of graphemeto-semantic and grapheme-to-phoneme mappings have been described in which intermediate representations not correspond to words (Hinton & Shallice, 1991; Plaut, McClelland, Seidenberg, & Patterson, 1996; Seidenberg & McClelland, 1989), leaving uncertain the theoretical need for whole-word codes in language processing (Besner, Twilley, McCann, & Seergobin, 1990; Coltheart, Curtis, Atkins, & Haller, 1993; Seidenberg & McClelland, 1990) With this brief exposition of a general language processing architecture, we now proceed to a discussion of the processing impairments underlying auditory comprehension and speaking disorders in Wernicke’s aphasia Jeffrey R Binder Input Phoneme 184 Output Phoneme Output Grapheme Semantic Object Feature Input Grapheme Figure 9.3 A language-processing architecture with intermediate representational levels (ovals) Unidirectional arrows show the typical directions in which information spreads during language tasks At a local level, however, these connections are probably bidrectional, allowing continuous interactions between adjacent representational levels Auditory Comprehension Disturbance Because comprehension of spoken words depends on the auditory system, speech comprehension deficits in Wernicke’s aphasia could be due to underlying abnormalities of auditory processing Luria, for example, theorized that speech comprehension deficits reflect an inability to discriminate subtle differences between similar speech sounds (Luria, 1966; Luria & Hutton, 1977) Although a discussion of acoustic phenomena in speech sounds is beyond the scope of this chapter, a few examples might serve to illustrate this point (the interested reader is referred to excellent reviews on this important and relatively neglected topic in clinical neuroscience: Klatt, 1989; Liberman, Cooper, Shankweiler, & Studdert-Kennedy, 1967; Oden & Massaro, 1978; Stevens & Blumstein, 1981) Speech contains both periodic sounds produced by vocal cord vibrations (exemplified by the vowels) and nonperiodic noises produced by turbulence at constriction points like the lips, teeth, and palate (exemplified by sounds like /s/ and /f/) The distribution of energy across the acoustic frequency spectrum (i.e., the relative loudness of low or high frequencies) at any point in time depends on the shape of the vocal tract (e.g., the position of the tongue, the shape of the lips, the position of the soft palate), which creates resonances that amplify or dampen particular frequencies Accentuated frequencies are referred to as formants; vowels are distinguished on the basis of the frequency position of the lowest three or four of these formants, which typically occupy frequencies in the range from 300 to 4000 Hz With rapid changes in vocal tract shape, such as those that occur during production of consonants like /b/ and /d/, the formants rapidly change position; this is referred to as formant transition One cue for distinguishing between consonants is the direction of movement (i.e., up, down, or straight) of these transitions In some consonants, such as /p/ and /t/, a very brief noise burst precedes the onset of vocal cord vibration Thus, /b/ and /p/, which are both produced by opening the lips and therefore have very similar formant transitions, are distinguished largely on the basis of this burst-to- Wernicke Aphasia periodicity onset asynchrony, referred to as voice onset time The inability to detect acoustic cues such as those distinguishing /b/ from /d/ or /b/ from /p/ might lead to misinterpretation of bay as day or bye as pie, for example, causing severe comprehension disturbance Because acoustic events in speech occur rapidly, other investigators have proposed an underlying problem with rapid processing in the auditory system, leading to the inability to discriminate phoneme order (e.g., hearing cast as cats or task as tax) or impaired perception specifically involving rapid dynamic phenomena such as formant transitions and differences in voice onset time (Brookshire, 1972; Efron, 1963; Tallal & Newcombe, 1978; Tallal & Piercy, 1973) The hypothesis that auditory processing deficits underlie the speech comprehension problem in Wernicke’s aphasia has been tested in several ways One task paradigm involves explicit identification or labeling of speech sounds For example, subjects hear a word or nonword (e.g., ba) and must select a matching visual word or nonword in an array containing phonologically similar items (e.g., BA, DA, PA) Patients with Wernicke’s aphasia perform poorly in such tests (Basso, Casati, & Vignolo, 1977; Blumstein, Cooper, Zurif, & Caramazza, 1977; Blumstein, Tartter, Nigro, & Statlender, 1984; Goldblum & Albert, 1972; Reidl & StuddertKennedy, 1985) It is critically important to note, however, that this type of task requires the integrity of two possibly distinct processes That is, the identification task not only requires auditory processing but also the ability to match the auditory percept to another, nonidentical stimulus (the visual form) In an effort to disentangle these components, investigators have employed sensory discrimination paradigms that not require such cross-modal association In a typical experiment of this type, the subject hears two speech sounds and must merely decide if these are identical or different Deficits in this discrimination task are much less pronounced than in the identification task, with some Wernicke aphasics performing within the normal range (Blumstein, Baker, & Goodglass, 1977; Blumstein et al., 1984; Reidl & Studdert-Kennedy, 1985) 185 Many patients tested with both paradigms are found to be deficient in the identification task, but not in the discrimination task, demonstrating the essentially independent nature of these deficits Most important, there does not appear to be a necessary correspondence between deficits in either of these tasks and measures of speech comprehension: Patients are found who show severe comprehension disturbances and normal discrimination, and others are found who have marked identification and discrimination deficits, but relatively normal comprehension (Basso et al., 1977; Blumstein et al., 1977; Blumstein et al., 1984; Jauhiainen & Nuutila, 1977; Miceli, Gainotti, Caltagirone, & Masullo, 1980) The fact that Wernicke aphasics often perform normally on phoneme discrimination tests even when they are unable to identify phonemes explicitly suggests that their speech comprehension deficit is unlikely to be due to impaired auditory processing alone Rather, the deficit elicited in these studies reflects an inability to use auditory information to access associated linguistic representations Having adequately perceived a speech sound, the Wernicke aphasic is typically unable to retrieve associated information, such as its written form, picture equivalent, or meaning A similar dissociation between sensory and associative processing in patients with fluent aphasia was documented by Faglioni et al using nonspeech auditory stimuli (Faglioni, Spinnler, & Vignolo, 1969), further illustrating the independence of comprehension deficits from auditory perception Patients with left hemisphere lesions in this study showed an intact ability to discriminate between two meaningless nonspeech sounds, but were impaired in a task requiring matching meaningful nonspeech sounds (animal noises, machine noises, etc.) to pictures Deficits in the latter task were significantly correlated with speech comprehension deficits as measured by the Token Test Recent research has further explored this difficulty in retrieving information associated with speech stimuli in Wernicke aphasia This problem could be explained in any of three ways: (1) as an impairment in activating the information, (2) as a Jeffrey R Binder loss or corruption of the information itself, or (3) as an impairment in using the information once it is activated These possible scenarios are not mutually exclusive, and in fact there is evidence supporting all three, suggesting that variable combinations of these deficits might occur in different patients Before embarking on an assessment of this evidence, it would be useful to review briefly some current ideas about how information associated with words and concepts might be organized and represented in the brain We store information about words internally as a result of encountering the words in various contexts throughout life This information collectively provides the meaning (or meanings, literal and figurative, verbal and nonverbal) of the word The study of word meaning is referred to as semantics, and the processes by which word meanings are stored, retrieved, and used are collectively called “semantic processes.” A great deal of theoretical and empirical work has expanded our conception of such processes since Wernicke articulated his simple notion of word meaning as a connection linking sensory memories of an object Most notable is the recognition that in addition to sensory attributes associated with objects, semantic processing concerns the learning and retrieval of conceptual categories and the hierarchical relationships between different categories To take a simple example, we learn by visualauditory association that an object with four legs of a certain length range, a squarish platform resting on the legs, and a panel rising from one end of the platform, is called CHAIR We learn that a chair has other typical sensory attributes such as being inanimate, quiet, and able to support weight We discover the functions of a chair by seeing it used and by using it ourselves The concept of CHAIR is said to be a basic-level concept, because all objects possessing these simple structural and functional attributes are similarly categorized as CHAIR (Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976) In addition to associating these direct sensory impressions with the word CHAIR, however, we learn about abstract attributes of chairs, such as the 186 fact that they are nonliving, often contain wood, and are made by people Using this information, we learn to associate chairs to varying degrees with other types of objects that share some of the same sensory, functional, or abstract attributes, resulting in the formation of hierarchical relationships between words Reference to these relationships enables the formation of superordinate categories that include objects with similar attributes For example, based on the knowledge that it is manmade, useful in a home, can be moved from place to place, and is not mechanical, CHAIR becomes a member of the superordinate category FURNITURE Other members of this category (e.g., TABLE, DESK, COUCH) differ from CHAIR in terms of specific sensory or functional attributes; these are the basic-level neighbors of CHAIR Finally, a large number of words become associated with CHAIR as a result of how chairs are used in daily life and in larger social contexts; these are the function associates of CHAIR The facts concerning where and how chairs are typically used, for example, create function associations between CHAIR and HOME, CHAIR and RELAX, and CHAIR and READ Facts concerning society and chairs create function associations between CHAIR and EXECUTION, CHAIR and COMMITTEE, and CHAIR and BARBER The sheer number and complexity of such relations stored in the human brain are staggering, and they are an essential base on which the comprehension and formulation of language depend Some studies suggest that this network of semantic representations is altered or defectively activated in Wernicke’s aphasia In most of these studies, patients were required to judge the degree of relatedness between words or pictures In an experiment by Zurif et al (Zurif, Caramazza, Myerson, & Galvin, 1974), for example, fluent aphasics, most of whom had mild Wernicke’s aphasia, were shown groups of three words and asked to pick the two that “go best together.” Unlike the nonfluent aphasia patients and normal control subjects, fluent aphasics showed very poorly defined categorization schemes, with maintenance of only the most Wernicke Aphasia broad category distinctions (e.g., human versus nonhuman) Goodglass and Baker (1976) presented subjects with a picture followed by a series of spoken words; for each word, the subjects indicated whether the word was related to the picture The types of word–picture relations tested included identity (e.g., ORANGE and picture of orange), sensory attribute (JUICY and picture of orange), function (EAT and picture of orange), superordinate category (FRUIT and picture of orange), basic-level neighbor (APPLE and picture of orange), and function associate (BREAKFAST and picture of orange) As anticipated, Wernicke aphasics performed poorly in this task relative to normal controls and aphasics with good comprehension In addition to this quantitative difference, however, qualitative differences were notable Subjects with good comprehension had relative difficulty recognizing the basic-level neighbor relations, while the Wernicke patients recognized (that is, responded affirmatively to) this type of relation more easily than other relations Unlike the other patients, Wernicke patients had particular difficulty recognizing function relations When performance in this task was compared with performance in confrontation naming of the same pictures, it was found that patients had more difficulty making relatedness judgments for items they were unable to name These performance patterns have been largely replicated in other studies (Chenery, Ingram, & Murdoch, 1990; McCleary, 1988; McCleary & Hirst, 1986) Several studies focused specifically on the integrity of superordinate–basic level relations Grossman (1980, 1981) used a task in which aphasics were given a superordinate category (e.g., FURNITURE) and had to generate basic-level examples of the category Responses were scored using published prototypicality ratings (Rosch et al., 1976) that indicate the degree to which an item is a typical or central example of the category (e.g., DESK is a central example of the category FURNITURE, while LAMP is a more peripheral member) Nonfluent aphasics produced exemplars with high prototypicality ratings, whereas fluent 187 aphasics produced more peripheral items and often violated the category boundaries altogether Grober et al (Grober, Perecman, Kellar, & Brown, 1980) assessed the integrity of superordinate category boundaries using a picture–word relatedness judgment task like that employed by Goodglass and Baker The degree of relatedness to the target category was manipulated so that word items included central members of the category, peripheral members, semantically related nonmembers (e.g., WINDOW for the category FURNITURE), and semantically unrelated nonmembers (e.g., HORSE for the category FURNITURE) Anterior aphasics accurately classified peripheral members and semantically related nonmembers, suggesting intact category boundaries, while Wernicke aphasics often misclassified these items, indicating impaired discrimination near category boundaries Similar conclusions were reached by Kudo (1987), who used a task in which patients judged whether depicted objects were members of hierarchical superordinate categories (domestic animal, beast, animal, and living thing) Aphasics showed abnormally diffuse categorization schemes, in that they frequently included semantically related nonmembers in categories (e.g., they included GIRAFFE as a domestic animal) This abnormality was strongest in severe fluent aphasics Despite this evidence, other investigators have questioned whether these findings necessarily indicate a defect in the structural organization of the semantic system itself Claims to the contrary are based on a series of studies measuring semantic priming effects during word recognition tasks Normal subjects take less time to decide if a stimulus is a word (the lexical decision task) if the stimulus is preceded by a semantically related word (Meyer & Schvaneveldt, 1971) The preceding word, or prime, is thought to activate semantic information shared by the two words, resulting in a partial spread of activation that lowers the recognition threshold for the second word (Collins & Loftus, 1975; Neely, 1977) Milberg and colleagues showed in several studies that patients with Wernicke’s aphasia demonstrate semantic priming Jeffrey R Binder effects that are as robust as those in normal persons (Blumstein, Milberg, & Shrier, 1982; Milberg & Blumstein, 1981; Milberg, Blumstein, & Dworetzky, 1987; Milberg, Blumstein, Saffran, Hourihan, & Brown, 1995) These basic results were replicated by several investigators (Chenery et al., 1990; Hagoort, 1993) Moreover, when the same patients were presented with word pairs and asked to judge explicitly whether the words were semantically related, they showed deficits like those observed in other studies using explicit semantic judgment tasks (Blumstein et al., 1982; Chenery et al., 1990) Thus, the patients showed normal semantic priming for word pairs, but they were impaired when asked to explicitly identify the semantic relationships that underlie the priming effect These findings have led a number of investigators to conclude that the network of stored semantic representations is largely intact in Wernicke’s aphasia, and that the deficit underlying the language comprehension and naming deficits shown by these patients consists of an inability to explicitly retrieve and manipulate this information The presence of semantic priming, however, does not necessarily indicate that semantic representations are intact The past two decades have witnessed the development of neural network models of semantic information retrieval that could explain preserved semantic priming even within a defective semantic system A full explanation of these models is beyond the scope of this chapter, and the reader is referred to several excellent reviews (Hinton, McClelland, & Rumelhart, 1986; Hinton & Shallice, 1991; McClelland & Rumelhart, 1986) In these models, perceptual features of words (i.e., letter shapes and phonemes) and semantic features associated with words are represented by large numbers of units in an interconnected network Connections exist between units of the same level (e.g., between representations of different phonemes) and between levels These connections may be excitatory or inhibitory, and the strength of each connection is defined by a numerical weight Through real-world experience, the network learns to associate combinations of graphemes or phone- 188 mes with appropriate semantic features by incremental adjustment of the connection weights Knowledge about words in such models is said to be distributed because the network can learn to correctly associate a large number of different words with a large number of semantic features using the same sets of units and connection weights It is the precise tuning of these excitatory and inhibitory weights that allows similar words like cat and cot to activate entirely different semantic features, very different words like cot and bed to activate very similar semantic features, and reliable behavioral distinctions to be made between words with very similar meanings Such networks exhibit many characteristic phenomena shown by human semantic systems, including automatic formation of categories and prototypes (pattern generalization), word frequency and context effects in recognition tasks, and semantic priming (Becker, Moscovitch, Behrmann, & Joordens, 1997; Masson, 1995; McClelland & Rumelhart, 1986; Moss, Hare, Day, & Tyler, 1994) When “lesioned” by random removal of units or connections, or by random changes in connection weights, such networks exhibit characteristic phenomena shown by neurological patients, including mixtures of correct and incorrect performance on different items (rather than absolute loss of function), a graded decrease in performance that is dependent on lesion extent, phonemic and semantic paraphasias, and semantic retrieval deficits that may be category specific (Farah & McClelland, 1991; Hinton & Shallice, 1991; McRae, de Sa, & Seidenberg, 1997; Plaut & Shallice, 1993) To gain an intuitive feeling for how semantic priming might be preserved after damage to such a network, consider the schematic example in figure 9.4, which is similar to one of the models tested by Plaut and Shallice (1993) Small boxes indicate individual featural units in the network Such units are not strictly analogous to individual neurons; rather, each unit can be thought of as representing a smaller module that is itself composed of interconnected units Thus, there is no implied one-toone relationship between features (such as “has four Wernicke Aphasia legs”) and individual neurons Units in the left column represent the phoneme or grapheme units that encode features of the word name, such as the graphemes C, A, and T for cat Units in the right column represent semantic features associated with the word, such as “eats mice” or “has four legs.” The middle column represents intermediate units, which provide the network with sufficient computational power to learn arbitrary mappings between patterns represented on the perceptual and semantic units.2 The lines between units represent excitatory connections (for display purposes, the connections between units of the same level are not shown) After suitable training accompanied by incremental adjustments of the weights on all connections, the presentation of any word pattern (e.g., CAT) to the perceptual layer at left results in activation of the correct associated semantic pattern (figure 9.4A) Semantic priming is due to the fact that the semantic pattern of a following semantically related word (e.g., DOG) partially overlaps that of the priming word (figure 9.4B), and this second pattern will be activated more easily because of residual activation from the priming word Lesioning of the network at any location results in a loss of precision in the mapping between perceptual and semantic patterns (figure 9.4C) Input words activate semantic patterns that resemble the target pattern, but with omissions or inappropriately included units As a result of this imprecision, words may be associated with incorrect semantic information and assigned incorrectly to superordinate categories; category boundaries and prototypes themselves lose definition, and subjects lack the precise information needed to judge semantic relatedness Because the activated semantic pattern partially resembles the target pattern, however, semantic priming, which is an imprecise phenomenon that depends only on partial overlap between semantic patterns, will be preserved because of the large number of semantic units contained within each pattern A number of other observations from studies of Wernicke’s aphasia are consistent with such an account Goodglass and Baker (1976) and Chenery 189 et al (1990) documented one such phenomenon during relatedness judgment tasks performed by normal persons and aphasics When asked to decide if two words were related, normal subjects and aphasics with good comprehension were slower to respond and made more errors when the two words were basic-level neighbors of the same superordinate category (e.g., CAT and DOG) than when the words were associated in other ways, such as object-superordinate (CAT and ANIMAL) or object-attribute (CAT and FURRY) In contrast, Wernicke aphasics paradoxically did not show this relative difficulty with basic-level neighbor relations This difference can be explained by a loss of distinctiveness of the semantic patterns activated by different basic-level neighbors In the normal state, basic-level neighbors such as CAT and DOG activate highly overlapping but distinguishable patterns of semantic units Inhibitory connections, which suppress activation of DOG semantic units when CAT is presented, for example, are particularly important in this regard, and allow normal subjects to respond appropriately when asked questions like “Does a cat growl at strangers?” In responding affirmatively to word pairs such as CAT-DOG during a relatedness judgment task, subjects must overcome this inhibition between basic-level neighbors In the lesioned state, such distinctions are blurred, and patterns of semantic unit activity for basic-level neighbors of the same category become more similar As a result, there is less inhibition between basic-level neighbors, and subjects with damage to the network paradoxically recognize the relationship between such neighbors more accurately and quickly than in the normal state Milberg et al documented another interesting phenomenon in studies of semantic priming using nonwords (Milberg, Blumstein, & Dworetzky, 1988b) As in previous semantic priming studies, the subjects were presented with stimulus pairs (in this case, auditory stimuli) and were asked to perform a lexical decision task on the second stimulus of the pair The initial phoneme of the first stimulus in the pair was manipulated to produce primes that were either semantically related words Jeffrey R Binder 190 A Grapheme B Intermediate A Semantic Grapheme Intermediate Semantic legs A legs B wings C D fins D fins E meows E meows F barks F barks G fur G fur B C wings scales scales O tail O tail P house pet P house pet Q sea animal Q sea animal R eats mice R eats mice chews bone S S T T chews bone Figure 9.4 A schematic representation of part of the pathway from input graphemes to semantics (pathway in figure 9.2), illustrating the effects of semantic priming and structural damage The diagram has been greatly simplified for the sake of clarity, in that the grapheme layer contains no representation of letter position; less than half of the possible connections are drawn; no representation of the connection strengths is given; and only a very small portion of the total set of semantic units is shown (A) Presentation of “cat” to the grapheme layer, represented in the left column by shading of the appropriate graphemes, produces patterns of activation in the intermediate and semantic units determined by the set of connection strengths between each of the layers, which have been adjusted through experience with cats and the letter sequence “cat.” (B) If presentation of “dog” quickly follows, as in a semantic priming experiment, activation of the semantic units appropriate to dogs is facilitated by residual activation from those units that were activated by “cat” and that are shared by dogs and cats (four legs, fur, tail, house pet), while activation of those features specific to cats (meows, eats mice) decays (C ) Damage to a portion of the network, represented here by the removal of several intermediate units and their connections, disrupts the pattern of activation input to semantic units, resulting in activation of inappropriate units (barks, scales) or failure to activate appropriate units (fur) The activation of an incorrect pattern of semantic units disrupts performance on semantic tasks, but semantic priming of “dog” may still be possible because of the preserved activation of a sufficient number of shared semantic units (four legs, tail, house pet) Semantic priming may even be exaggerated in some cases if incorrect activation results in falsely “shared” features (barks) Wernicke Aphasia 191 C Grapheme Intermediate Semantic legs A B C wings D fins E meows F barks G fur scales O tail P house pet Q sea animal R eats mice S T chews bone Figure 9.4 Continued (e.g., CAT before DOG), nonwords differing from the semantic prime by one phonetic feature (GAT before DOG), or nonwords differing from the semantic prime by two phonetic features (WAT before DOG) The baseline condition used unrelated primes (NURSE before DOG) Unlike nonfluent aphasics, who showed priming effects only for the undistorted real-word prime, fluent aphasics showed priming effects for all phonetically distorted nonword conditions relative to the unrelated word baseline These results suggest that in fluent aphasia, nonwords more easily activate semantic patterns asso- ciated with phonetically similar real words To understand this phenomenon, recall that in the normal state the semantic network is able to accurately distinguish phonetically similar words such as CAT and COT and to associate each with an appropriate pattern of activation on the semantic units This feat is accomplished despite the fact that because the phonetic inputs for CAT and COT are similar, the initial activity across the set of semantic units is relatively similar after presentation of CAT and COT (figure 9.5) The separation of CAT and COT is possible because of recurrent interactions between units in the network, which cause the semantic units to gradually settle into a steady state that is very different for CAT and COT (figure 9.5A) Networks that behave in this way are known as attractor networks, and the patterns toward which the units gradually settle (the black dots in figure 9.5) are the attractor states Just as perceptually similar words like CAT and COT move gradually toward different attractor states, nonwords that are perceptually similar to words may move toward the attractor states for those words, resulting in partial activation of the semantic pattern of the word (figure 9.5) In normal subjects, this phenomonon depends on the degree of perceptual similarity between the nonword and the word associated with the attractor state; nonwords that differ by a greater number of phonetic features are less likely to move toward the attractor state (Milberg, Blumstein, & Dworetzky, 1988a) After the network is lesioned, the area of semantic space dominated by a given attractor state (called the “attractor basin”) becomes distorted and less sharply defined (Hinton & Shallice, 1991), with the result that activation patterns elicited by words and nonwords are more likely to move toward attractor states of phonetically similar words, resulting in enhanced semantic priming of these words (figure 9.5B) The general effect of such lesions is thus to blur the distinctions between words and phonetically related (or graphemically related) nonwords This loosening of phoneme-to-semantic mapping may also explain the observation by Blumstein et al that patients with Wernicke aphasia not show the usual lexical Jeffrey R Binder 192 A Phonetic B Semantic cat GAT • Phonetic Semantic cat GAT CAT CAT COT • COT BED bed • BED • cot bed • • cot Figure 9.5 The role of attractor states in phoneme-to-semantic (or grapheme-to-semantic) mapping The box on the left of each figure represents spoken word or nonword input The larger box on the right represents semantic space Points in this semantic space represent patterns of activation across a set of semantic units (activation states) Three such states are marked by black dots and correspond to the concepts cat, cot, and bed Lines and arrows show the initial state of the semantic network when it is presented with a given input and subsequent changes as the network settles into an attractor state Shaded regions are the attractor basins for each attractor state Any input that initially produces an activation state that falls within an attractor basin will eventually reach the attractor state for that basin (A) In the normal state, attractor dynamics allow similar inputs, such as cat and cot, which produce similar initial activation states, to eventually activate very different states in semantic space Conversely, very different inputs, such as cot and bed, may nevertheless reach relatively similar states in semantic space (B) Damage to the network causes distortion and loss of definition of the attractor basins As a result, semantic states resulting from a given input word may gravitate toward incorrect (phonologically or semantically related) attractors (cot Ỉ cat, bed Ỉ cot), and attractor states may be reached more easily from nonword inputs (Based on Hinton & Shallice, 1991.) effect on placement of perceptual boundaries during phoneme categorization (Blumstein, Burton, Baum, Waldstein, & Katz, 1994) In summary, the speech comprehension disturbance in Wernicke aphasia is not well explained by a phoneme perceptual disturbance The input phoneme representations in figure 9.3 appear to be, for the most part, intact In contrast, there is a deficit either in the pathway from input phonemes to semantics or within semantic representations (or both), as demonstrated by the inability of Wernicke aphasics to match perceived phonemes with their associated visual forms or meanings Several lines of evidence suggest that semantic representations are activated inaccurately, causing blurring of category boundaries, loss of distinctiveness between basic-level neighbors, inability to judge semantic relatedness, and abnormal activation of semantic representations by wordlike nonwords Although preserved semantic priming in Wernicke’s aphasia has been interpreted as indicating intact semantic representations, an alternative explanation is that Wernicke Aphasia priming merely reflects partial overlap of semantic activation between word pairs and does not require that this activation be precise or accurate Paraphasia As described earlier, Wernicke aphasics produce a variety of speech output errors involving sound elements within words (phonemic paraphasia), grammatical units such as word stems or suffixes (morphemic paraphasia), and whole words (verbal paraphasia) At least since Freud’s claim that “paraphasia in aphasic patients does not differ from the incorrect use and distortion of words which the healthy person can observe in himself in states of fatigue or divided attention” (Freud, 1891/1953, p 13), it has been recognized that speech errors made by aphasic patients share many features with those made by normal speakers (Blumstein, 1973; Buckingham, 1980; Dell et al., 1997; Garrett, 1984; Schwartz, Saffran, Bloch, & Dell, 1994) In recent years, quantitative techniques and theoretical models arising from the study of normal “slips of the tongue” have been applied productively to the analysis of speech errors made by aphasics This section briefly recounts some of the important findings from these studies as well as a computational model that explains many of the basic paraphasic phenomena exhibited by fluent aphasics It is clear that paraphasic errors are not entirely random One example of a general rule operating at the phoneme level is the frequent occurrence of contextual errors—phoneme errors that are influenced by other nearby phonemes Examples include anticipations, in which a later sound is duplicated in an earlier utterance (“park bench” Ỉ “bark bench”); perseverations, in which an earlier sound is duplicated in a later utterance (“beef noodle” Ỉ “beef needle”); and exchanges, in which two sounds exchange places (“big deal” Ỉ “dig beal”) Contextual errors are the principal type of phoneme error in normal slips of the tongue (Nooteboom, 1969), and imply a speech production mechanism in which the selection of each phoneme to be uttered is partly influenced by preceding and following 193 phonemes A related finding is that phonemes interacting in this way tend to be those that are similar to each other That is, phonemes are more likely to be switched with other phonemes if they share similar phonetic features (e.g., /b/ and /d/ share the same manner and voicing features) and if they occupy the same position within their respective syllables (e.g., the /b/ and /d/ in “big deal” are both syllable onset phonemes) (Blumstein, 1973; Lecour & Rouillon, 1976; MacKay, 1970; ShattuckHufnagel & Klatt, 1979; Stemberger, 1982) Thus, the mechanism that selects phonemes appears to be influenced by other surrounding phonemes, particularly if these are easily confused with the target phoneme Analogous contextual phenomena are observed for words within multiword phrases Thus, there occur contextual word anticipations (“The sun is in the sky” Ỉ “The sky is in the sky”), perseverations (“The boy is reaching for the cookies” Æ “The boy is reaching for the boy”), and exchanges (“writing a letter to my mother” Ỉ “writing a mother to my letter”) in both aphasic and normal speech (Dell & Reich, 1981; Fromkin, 1971; Garrett, 1975; Lecour & Rouillon, 1976) Analogous to the confusability effects seen with phoneme errors, word substitutions show effects of semantic and grammatical class similarity That is, for both normal and aphasic speakers, substituted words are more likely to be semantically related to the target word (Buckingham & Rekart, 1979; Dell & Reich, 1981; Fay & Cutler, 1977; Fromkin, 1971; Garrett, 1992) and are more likely to be from the same grammatical class (i.e., noun, verb, adjective) as the target word (Fay & Cutler, 1977; Gagnon, Schwartz, Martin, Dell, & Saffran, 1997; Garrett, 1975) than would be expected by chance alone Thus, the mechanism used for selecting words appears to be influenced by other nearby words in the planned utterance, particularly those that are from the same grammatical class as the target word, and by the possible “semantic neighbors” of the target word These examples involving phoneme and wordlevel substitutions by no means capture all of the error patterns observed in fluent aphasia Of Jeffrey R Binder particular theoretical interest are errors that appear to indicate interactions between word and phoneme information One example of this is the formal paraphasia, a real-word error that is phonologically but not semantically similar to the intended target word (“horse” Ỉ “house”) Although formal paraphasias are real words, the question of whether they represent phoneme or word-level errors has been controversial, for several reasons First, these errors are not common in most patients Second, patients who produce formal paraphasias also utter nonwords (neologisms) that are phonologically related to targets These data have usually been accepted as evidence that formal paraphasias represent phoneme-level errors that happen by chance to result in real words (Buckingham, 1980; Butterworth, 1979; Ellis, Miller, & Sin, 1983; Lecour & Rouillon, 1976; Nickels & Howard, 1995) In contrast, several investigators have recently provided evidence for a greater-than-chance incidence of formal paraphasias in some aphasics That is, in producing errors that sound similar to the intended target, some patients appear to produce real words (as opposed to nonwords) at a higher rate than would be predicted by chance (Best, 1996; Blanken, 1990; Gagnon et al., 1997; Martin, Dell, Saffran, & Schwartz, 1994) The same phenomenon has been observed in studies of normal slips of the tongue, which typically show a higher than chance rate of word compared with nonword errors (Baars, Motley, & MacKay, 1975; Dell & Reich, 1981) If these errors truly represent incorrect word selection rather than phoneme errors that happen to have resulted in words, we might expect the errors to be in the same grammatical word class as the intended target Evidence suggests that this is the case (Dell et al., 1997; Fay & Cutler, 1977; Gagnon et al., 1997) These findings are important because they suggest a production mechanism in which selection at the word level is partly constrained by information about the sound of the word, possibly through feedback from phoneme to word-level representations Other evidence for interaction between phoneme and word information during speech production 194 comes from the observation of mixed paraphasias, in which a real-word error response is related both phonologically and semantically to the intended target (“skirt” Æ “shirt”) While such errors comprise only a small proportion of the total errors made by aphasic patients, the issue again is whether this proportion is small enough to be explained as coincidence Studies of both normal and fluent aphasic subjects show that the incidence of mixed errors is significantly greater than would be expected by chance alone (Dell & Reich, 1981; Dell et al., 1997; Harley, 1984; Martin, Gagnon, Schwartz, Dell, & Saffran, 1996) As with formal paraphasias, the higher than expected incidence of mixed paraphasias suggests that phonological resemblance to a target word is somehow enhancing the selection of an error word In the case of the mixed error, this phonological resemblance between target and error words is acting in concert with a semantic resemblance Many of the basic error phenomena observed in normal slips of the tongue and aphasic paraphasia—contextual effects, similarity effects, and phoneme–word interactions—can be seen as the natural product of a neural network in which word selection and phoneme selection partly overlap in time and influence each other through an interactive spreading activation mechanism Models of this kind have been presented by Dell and colleagues (Dell, 1986; Dell & O’Seaghdha, 1992; Dell et al., 1997) and by others (Harley, 1984; Roelofs, 1992; Stemberger, 1985) Figure 9.6 shows a simplified diagram of a hypothetical portion of such a network A central starting point of the model is that speech production involves two distinct processes or stages of information access that partly overlap in time The first of these is a translation from the abstract concept the speaker wishes to express (the semantic representation) to a word or ordered string of words that express the concept This process is referred to as lemma access “Lemma” refers to a type of representation in the brain similar to a word, but with some important differences.3 The lemma representation of a word contains information about its syntactic role (noun, verb, etc.), and the lemma Wernicke Aphasia Semantic Lemma Phoneme (onset, vowel, coda) Phonetic Feature 195 tail fur edible hen h aspirated living rat r k voiced vocal cat p ch labial hat b nasal inside legs pen e alveolar heavy pear ae manmade chair n velar wood bear t sonorant for rest bed r stop d fricative Figure 9.6 Schematic representation of an interactive, spreading activation model of speech production The network is shown at a moment in time during production of cat A set of semantic units have produced activation of the target lemma as well as several semantically related lemmas (rat, bear) Position-specific (onset, vowel, and coda) phonemes are activated as a result of the spread of activation from the lemma to the phoneme level The network has just selected the onset phoneme /k/, resulting in increased activation of the phonetic feature nodes (aspirated, velar, stop) associated with /k/ Note weak activation of the lemma hat despite lack of input from the semantic level, owing to feedback from the phoneme units for /ae/ and /t/ This feedback is postulated to be the main source of formal (i.e., semantically unrelated but phonologically related) paraphasic errors The lemma rat also receives phoneme-level feedback, and is more strongly activated than hat because of combined input from semantic and phoneme levels, increasing the likelihood of a mixed (semantic + phonological) error Finally, note that activation of the phonetic feature nodes, aspirated and stop, feed back to phoneme nodes, such as /t/, that share these features, increasing the likelihood of selection errors involving phonemes similar to the target phoneme Jeffrey R Binder 196 is a more abstract entity than a word in that it contains no information about the sound of the word, its phonological representation (Dell, 1986; Dell & O’Seaghdha, 1992; Kempen & Huijbers, 1983; Levelt et al., 1991) In this sense, the lemma differs from Wernicke’s notion of a word-sound image— a kind of memory of a word stored in auditory format—and it differs from the phonological lexicon found in many contemporary models of language processing (Allport & Funnell, 1981; Morton & Patterson, 1980), which is composed of phonological word forms In the second stage of the two-stage model, the sounds of the word are computed by translating the lemma into a string of ordered phonemes, a process referred to as phonological access As shown schematically in figure 9.6, the component nodes of the network are organized into semantic, lemma, phoneme, and phonetic feature levels.4 Each level contains a large number of nodes that represent, in a distributed manner, the individual’s fund of information about concepts, words, and phonemes Connections between nodes in adjacent levels represent relationships between concepts, words, phonemes, and phonetic features, which have been learned over time as a result of experience A key feature of the model is that connections between adjacent levels are reciprocal, permitting activation to flow in both top-down and bottom-up directions As with other models of this type, the activation state of each node in the network is computed at discrete points in time as a weighted sum of all the inputs to the node, plus the activation level at the immediately preceding point in time modified by a decay term, e.g.: Ai (t ) = Ai (t − 1) × (1 − d ) + ∑ a j (t − 1) × wij + noise j where Ai(t) is the activation of node i at the current time step t, Ai(t - 1) is the activation at the immediately preceding time point, d is a decay parameter, aj(t - 1) is the activation at the immediately preceding time point of a node j sending input to node i, and wij is the strength, or weight, of the connection between the sending node j and receiving node i Speech production in the model begins with activation of semantic nodes representing the concept that the subject wishes to express Semantic nodes are usually envisioned as representations of physical, functional, or associative properties (e.g., “has fur”) that collectively define a word, although the specific format in which semantic information is encoded is probably not critical for the model The active semantic nodes then send activation to all lemma nodes to which they are connected An important point is that a given semantic node is connected to all lemmas that share that semantic feature; thus activation of a given semantic node results in some degree of activation of all lemmas to which it is connected Lemma nodes, in turn, send activation to all phoneme nodes to which they are connected, and these phoneme nodes send activation on to all phonetic feature nodes to which they are connected Because of the reciprocal connections between layers, activation is also returned from the lemma to semantic level, from the phoneme to lemma level, and from the phonetic feature to phoneme level Similarity effects—the occurrence of word errors that resemble the target semantically (semantic paraphasia) and phoneme errors that resemble the target phoneme—are readily explained by such a model Because semantic nodes send activation to all lemmas to which they are connected, lemmas that strongly resemble the target will be activated nearly as much as the target itself For example, because chair and couch share many semantic features, activation of the semantic representation for chair will necessarily activate the lemmas of both chair and couch (and many other related items) Under normal conditions, the network is able to select the correct lemma on the basis of its activation being slightly higher than that of its semantic neighbors (adding to the model inhibitory connections between lemma nodes also helps to suppress activation of these neighbors), but adding noise to the system by partial damage to nodes or to connections can easily cause this fidelity to be compromised, resulting in lemma selection errors of a semantic nature Wernicke Aphasia Phoneme similarity effects happen by a similar mechanism, but are due to feedback from the phonetic feature level Because phonetic features have reciprocal connections to all phonemes that share that feature, activation of a given set of phonetic features by a phoneme node will cause reciprocal activation of other phoneme nodes that share those features Again, adding noise to the system can occasionally cause one of these phonetic neighbors to become more active than the target phoneme, resulting in a phoneme selection error on the basis of similarity Interactive effects—the occurrence of formal paraphasias and mixed semantic-phonological errors—are also a natural consequence of the model Recall that a higher-than-chance incidence of formal paraphasias means that errors that resemble the target phonologically are more likely than chance to form real words In the network model, this phenomenon is accounted for by feedback from phoneme to lemma levels When the target lemma (e.g., train) becomes activated, this produces activation of the phoneme nodes connected to the target lemma (e.g., /t/, /r/, /e/ and /n/) Reciprocal connections allow these phoneme nodes to feed back on all other lemmas to which they are connected, producing particularly significant activation of lemma nodes that share several phonemes in common with the target lemma (e.g., crane and trait) If selected, such a lemma will in turn increase the activation level of its phonemes, increasing their likelihood of being selected Although small, these effects increase the likelihood that a phonological neighbor of the target lemma will be produced, rather than a randomly generated nonword (e.g., prain) Mixed errors have a similar explanation, except that here the error results from a combination of semantic and phoneme influences at the lemma level That is, shared semantic features result in the activation of semantic neighbors of the target lemma, while shared phonemes cause the activation of phonological neighbors These influences add together to increase the likelihood that a mixed semanticphonological neighbor of the target, if one exists (e.g., plane), will be selected at the lemma level 197 Because contextual errors (anticipations, perseverations, exchanges) typically involve words that are near each other in time, an account of these phenomena requires a look at how the model handles multiword utterances A complete description is beyond the scope of this review, but the main point is that words within multiword sequences are, to a large degree, selected in parallel That is, as activation is accumulating in the lemma and phoneme nodes related to the first word in the string, activation also begins to accumulate in the nodes pertaining to the second word After reaching maximum levels, activation also takes time to decay back to baseline levels; there is thus residual activation in the nodes for a preceding word even as the nodes for a following word are being selected Moreover, because selection at the lemma level occurs earlier than selection at the phoneme level, activation at the lemma level for a following word may be occurring almost simultaneously with activation at the phoneme level for a preceding word This considerable temporal overlap occasionally creates selection of a lemma or phoneme node that is actually a target lemma or phoneme for a preceding or following word, resulting in contextual errors Specifically, anticipations are due to selection of a phoneme or lemma from a later word, which happened by chance to have been more activated than the target phoneme or lemma Perseverations are due to selection of a phoneme or lemma from an earlier word, which happened by chance to have been more activated than the target phoneme or lemma Exchanges are believed to reflect a mechanism that transiently suppresses the activation of a node after it has been selected For example, during phonological translation of the lemma for cat, the network transiently suppresses or inhibits the activation level of the phoneme node for /k/ after this is selected Although the mechanism by which this occurs is not clear, some sort of suppression appears necessary to prevent, for example, the /k/ phoneme from being chosen again and again for subsequent phoneme positions Exchanges thus occur when an anticipation error causes a phoneme from a following word to be selected prematurely, and the ... neuropsychological test of neuroimaging findings Proceedings of the National Academy of Sciences U.S.A., 95, 15855–15860 Von Stockert, T R (1 974 ) Aphasia sine aphasie Brain and Language, 1, 277 –282 174 Wernicke... (Basso et al., 1 977 ; Blumstein et al., 1 977 ; Blumstein et al., 1984; Jauhiainen & Nuutila, 1 977 ; Miceli, Gainotti, Caltagirone, & Masullo, 1980) The fact that Wernicke aphasics often perform normally... S (19 67) Towards the mechanism of “dynamic aphasia” Acta Neurologica Psychiatrica Belgica, 67, 1045–10 67 Masdeu, J C., Schoene, W C., & Funkenstein, H (1 978 ) Aphasia following infarction of the